Having daughters rather than sons makes you more liberal: a link followed by a plea

Andrew J. Oswald and Nattavudh Powdthavee write:

In remarkable research, the sociologist Rebecca Warner and the economist Ebonya Washington have shown that the gender of a person’s children seems to influence the attitudes and actions of the parent.

Warner (1991) and Warner and Steel (1999) study American and Canadian mothers and fathers. The authors’ key finding is that support for policies designed to address gender equity is greater among parents with daughters. This result emerges particularly strongly for fathers. Because parents invest a significant amount of themselves in their children, the authors argue, the anticipated and actual struggles that offspring face, and the public policies that tackle those, matter to those parents. . . The authors demonstrate that people who parent only daughters are more likely to hold feminist views (for example, to favor affirmative action).

By collecting data on the voting records of US congressmen, Washington (2004) is able to go beyond this. She provides persuasive evidence that congressmen with female children tend to vote liberally on reproductive rights issues such as teen access to contraceptives. In a revision, Washington (2008) argues for a wider result, namely, that the congressmen vote more liberally on a range of issues such as working families flexibility and tax-free education.

Our [Oswald and Powdthavee’s] aim in this paper is to argue, with nationally representative random samples of men and women, that these results generalize to voting for entire political parties. We document evidence that having daughters leads people to be more sympathetic to left-wing parties. Giving birth to sons, by contrast, seems to make people more likely to vote for a right-wing party. Our data, which are primarily from Great Britain, are longitudinal. We also report corroborative results for a German panel. Access to longitudinal information gives us the opportunity — one denied to previous researchers — to observe people both before and after they have a new child of any particular gender. We can thereby test for political ‘switching’. Although
panel data cannot resolve every difficulty of establishing cause-and-effect relationships, they allow sharper testing than can simple cross-section data.

They addressed the concerns about the research I’d expressed earlier.

Just one thing . . .

I have only one request, and I know it’s too late because the article is already scheduled to appear in a journal, but I’ll ask anyway. The article has lots of graphs and lots of tables–and I’ll spare you my detailed thoughts on these, because, again, it’s already scheduled to appear.

But one thing that I didn’t see graphed is what I would think is the most natural and important thing to graph: the estimated change in the probability of voting for the conservative party, comparing a parent of a boy compared to the parent of a girl. That is, the estimated effect on the vote of having a boy, compared to a girl. I assume this effect varies by sex and age of parent and also by age, number of previous children, past voting patterns, and other factors.

Earlier today we had a brief discussion here about interactions, and this is a great example for thinking about modeling and displaying them.

(The graphs that are in the Oswald and Powdthavee article give average numbers of boys and girls for voters of different parties, but that’s not quite what I’m looking for. As the authors so clearly explained, the key question is the effect of the sex of the child on parents’ attitudes and behavior, and I’d like a graph that would really show this. As it is, I honestly have difficulty figuring out the estimated effect size here. Yes, it’s great to see that coefficients are statistically significant–but I want to see what’s going on here. I want to see the estimated effect.)

11 thoughts on “Having daughters rather than sons makes you more liberal: a link followed by a plea

  1. the estimated change in the probability of voting for the conservative party, comparing a parent of a boy compared to the parent of a girl.

    I think more interesting would be the effect of having a child of either sex vs. no child at all. Do people (men?) begin more right wing, and become less so with the birth of a daughter? Or do they begin somewhere in the middle, moving right with the birth of a boy, left with the birth of a girl? Is one daughter all it takes to make someone more liberal, or does each additional daughter have an incremental effect? Etc., etc.

  2. Or, Jon, does that support the hypothesis? If voters tend to become more conservative as they get older [citation needed], then a lack of movement is a relative movement to the left.

  3. The data used in this study aren't a representative sample of households in Great Britain. Waves 1 to 10 were carried out in Great Britain, and waves 11 to 16 in the United Kingdom. These are not synonyms. One would have wished that the authors, as residents in the UK, would have known the difference.

  4. Alex,

    A flippant (but true) response is that changing the reference population from GB to UK represents an increase of a mere 2.5% or so — NI might be symbolically significant but numerically it barely exists.

    A more serious response is that it depends on what weights they use. If they use the standard BHPS longitudinal weights, the reference population is explicitly the 1991 GB population (later entrants get weighted to zero).

  5. I guess my own parents are anomalies then. I have 5 sisters (there are 6 of us total – no boys) and my parents are both extremely conservative… as are all of us daughters. My parents are not liberal in any sense of the word.

  6. Another possible interpretation, and one that I have not seen brought up yet, is that the link goes in the other direction: namely that the attitudes and actions of the parent may influence the gender of a person's children. I've read studies that show that engineers are more likely to have boys, and people who are more caregivers, like nurses, more daughters. Just google "engineers more sons". So maybe ones political inclinations does the same thing???

  7. Rene: Your parents are hardly anomalies. The effects are small and represent differences in averages, without much predictive power at the individual level.

    Hispanic Pundit: No, there is no evidence that engineers are more likely to have sons, and it is unfortunate that this mistaken finding got so much press! The article that purported to show this had statistical errors. See here or search Gelman Kanazawa on Google Scholar or search Kanazawa on this blog for more information on the topic.

  8. I wonder how this would apply to other societies. Imagine China, where there's a massive imbalance caused by the desire for sons – and thus the abortion of female fetuses (used to be killing the born girls). Would the prevalence of males tend to support the authoritarian regime? Would having girls affect, meaning improve, the integration of Muslims into Western Europe, as they move away from a male-dominated society?

  9. Andrew: I have no access to the paper you referred to, but I did find your paper "Of beauty, sex, and power:Statistical challenges in estimating small effects" from your website.

    In the paper, the correlation is not significant when attractiveness (A) is rescaled and regressed onto proportion of girl birth (G).

    I am not sure whether such an inconsistency would raise concerns for the original test by Kanazawa. I do like the rest of the argument in the paper.

    1 different hypotheses were tested.
    1.1 Kanazawa: there is difference in G between the highest level vs. the rest 4 categories of A.
    1.2 Regression: there is a linear trend between A and G across the levels of the A, or as you pointed out 'The new scale makes it comparable to a binary variable contrasting cases that rank above and below
    average'

    2 The regression analysis seems to only use 5 observations (right?). I am not sure how the test on the regression coefficient performs with such a small sample size. I would have more trust on the p-value from a contingency table including many more observations and relying on fewer assumptions.

Comments are closed.